iDerive Analytics Platform

Holdout-market testing is the only clean answer. Pause Amazon ad spend in a geographically controlled holdout market for 3–4 weeks while running normally elsewhere, and measure DTC lift in the control versus holdout. AMC's cross-channel overlap reports help but require DSP plus a non-trivial identity stitch between Amazon shoppers and DTC CRM. Anything short of holdout testing is correlation, not causation, and correlation gets you a 70% overattributed halo number that justifies any budget decision you want. Most brands we work with discover actual halo is 15–25% of what their previous agency was claiming.

Start by modeling contribution margin over the repurchase window, not the first order. If your LTV looks like $180 revenue against $45 ad spend over 18 months for a supplement brand, your LTV-adjusted break-even TACOS is 25%, not the 15% that first-order accounting would suggest. The catch is that Amazon's closed-garden attribution doesn't show repurchase cleanly. You need Subscribe & Save data plus brand-level order history to model it, and most agencies don't build the view. This is where AMC plus a warehouse pipe (iDerive for us, something similar elsewhere) earns back its cost inside of a year.

Three tiers. Efficiency: ACOS, TACOS, cost per new-to-brand customer, ad-attributed conversion rate. Growth: unit velocity, share of category search, new-ASIN ramp time, organic rank movement on category keywords. Strategic: share of voice against three named competitors, AMC-measured incrementality, Subscribe & Save enrollment rate. ACOS alone incentivizes brand-term harvesting — cheap clicks from people who were going to buy anyway — and punishes category-growth bets. A good KPI tree holds the agency accountable to growth without letting them hide behind a single metric.

Subscribe & Save orders are roughly 60–75% non-incremental. Those shoppers would have repurchased via Subscribe & Save or direct search without your ads. That matters because a 25% ACOS looks very different when only 30% of ad-attributed sales are truly new. The real number you want is ACOS on new-to-brand customers, which AMC exposes cleanly. Most pet brands are overpaying for ads by 20–30% because their ACOS math bakes in loyal-customer rebuys they didn't need to pay for.

When you’ve outgrown pre-built reports, meaning your questions no longer match any canned dashboard. Helium 10 and similar tools are rules-based: they tell you what’s happening against known benchmarks. AMC is query-based: it answers questions no one’s pre-built for you ("what’s my path-to-purchase overlap between Sponsored Products and DSP for new-to-brand customers who came from competitor-conquest keywords?"). The trigger for graduating is usually (a) DSP spend above $50K/month, (b) cross-ad-type path analysis Sponsored Ads reports can’t produce, or (c) stitching Amazon behavior into a larger data model across DTC, retail-media networks, and CRM. Under those conditions, AMC typically cuts 15–25% of ad waste inside two quarters. Outside them — a $2M brand running Sponsored Products only — AMC is overkill and a dashboard tool tells you what you need.

They're measuring different things. AMC's NTB is "first-time purchaser of your brand on Amazon within the trailing 12 months". Amazon-specific. Triple Whale/Northbeam NTB is "first-time purchaser on your DTC site within some window", DTC-specific. A shopper who bought on Amazon last year and on DTC today is NTB to Triple Whale but not to AMC. To truly reconcile, you need identity resolution. Match Amazon shoppers (via AMC audience outputs if you have DSP, or via Subscribe & Save email fields) to your DTC CRM, then dedupe. Most brands don't close this loop and end up with two separate NTB numbers that both feel "right" from their respective tools. The unified number is almost always lower than either.

The instance is provisioned to your Amazon Advertising account, not the agency's, so legally, yes, the data belongs to you. Practically, the value depends on whether the SQL queries, data models, and custom audiences built inside the instance were documented and version-controlled. Most agencies don't document AMC work — it lives in their analysts' heads — so when you leave, you retain the raw data access but lose the institutional knowledge of what to query and why. Before signing an AMC-involved engagement, require that all queries, audience definitions, and dashboards be stored in a shared repository you control (Git, Notion, Google Drive, anywhere except the agency's internal wiki).